Economy and Turnout: Class Differences in the 2000 U.S. Presidential Election Uisoon Kwon University of Minnesota Duluth

Transcription

1 Economy and Turnout: Class Dfferences n the 2 U.S. Presdental Electon Usoon Kwon Unversty of Mnnesota Duluth For many years, scholars have been examnng the role of the economy on voter turnout. Does the economy matter? If t does, how wll the economy affect voter turnout and who does t affect more? I address these ssues by examnng four states durng the 2 U.S. presdental electon. Theoretcally, t s argued that the effect of economc adversty depends upon the degree of economy securty; therefore, the lower class votes at a lesser degree than the non-lower class. Based upon ths premse, I explore two hypotheses: () The voter turnout of the lower class wll be less than that of the non-lower class; and (2) the responsveness of lower class turnout to macroeconomc condtons wll be dfferent from the non-lower class. Furthermore, the nonlower class wll be demoblzed to a greater degree than the lower class when macroeconomc condtons are poor. I test these hypotheses usng a relatvely new method, ecologcal nference. Amercan ctzens have fought to gan votng rghts startng from the extenson of suffrage to property-less whte males n 84. Women dd not secure the vote untl the ratfcaton of the Nneteenth Amendment n 92 and Afrcan Amercans dd not have the full rght to vote untl Votng Rghts Act of 965. Despte the hstorcal struggle for unversal suffrage, low voter turnout n presdental electons remans a persstent problem n Amercan poltcs. When Bll Clnton was reelected presdent n 996, only 49 percent of the elgble electorate cast ballots. Snce the Clnton-Dole race n 996 was one-sded from the start, many expected turnout would rse n 2. However, despte beng the tghtest contest snce the 96 presdental electon between Kennedy and Nxon, a mere 5 percent voted n 2. Ths was a large dfference from the Kennedy-Nxon race when 63 percent of votng age ndvduals voted. Wth the slght excepton of the 992 electon when an economc recesson and Ross Perot s race as a thrd party canddate ncreased turnout by fve percent, the presdental electons were consstent wth lower turnout over tme. Although turnout n the 24 electon was the hghest snce 952, t was merely close to that of 992 electon and stll far from turnout rates durng the 96s. Furthermore, contemporary turnout rates have yet to reach that of the 9 th century; between 88 and 896, voter turnout rates ranged from 79.2 percent to 84. percent n the northern states. Even n the southern states, the average turnout rate durng ths tme perod was 6.3 percent (Kornbluh, 2). Ths artcle examnes two specfc questons. Frst, how has the lower class voted compared to the nonlower class? Second, to what degree has ther turnout been nfluenced by contextual factors when compared to the non-lower class? Despte the wdespread agreement that class matters, there s substantal uncertanty about the mpact of class on votng behavor when nfluenced by poor economc condtons. In ths paper I estmate the probablty of a vote usng a Bayesan herarchcal ecologcal nference model wth a covarate and examne the contextual effects of economy on turnout among the lower and non-lower classes. These results were used to test the hypothess that lower class turnout was stmulated by macro economc condtons to a lesser degree than was non lower class turnout. CLASS AND TURNOUT The study of ndvdual characterstcs as determnng factors n votng turnout emerged wdely wth the advent of votng surveys. The Natonal Electon Study (NES) and other surveys such the Census Populaton Survey, made t possble for scholars to fnd emprcal correlates of votng (Boyd, 98; Cassel & Hll, 98; Campbell, Converse, Mller & Stokes, 96; Mlbrath & Goel, 977; Shaffer, 98; Texera, 987; Wolfnger & Rosenstone, 98). Based on the result of these surveys, researchers have generally found that that ndvdual soco-economc status was correlated wth votng turnout. Indvduals who have more educaton and hgher ncomes tend to vote at hgher rates than those wth less educaton and lower ncomes. Accordng to Leghley and Nagler (992), lower class people vote at roughly 6 percent of the rate of upper class people. It seems to be obvous that ndvdual soco-economc status matters. If there s a sgnfcant relatonshp between turnout and soco-economc status, then what socal group or Indana Journal of Poltcal Scence, Wnter 27, 69

2 class contrbuted the most to the declne of votng turnout n Amercan poltcs? How and to what extent does ndvdual soco-economc status matter? Early research generally focused on the dfferences n voter turnout rates between lower class and hgher class ndvduals. (Burnham, 987; Reter, 979; Bennett, 99; Leghley & Nagler, 992; Patterson, 22). On the other hand, more recent lterature focuses on the queston of why ths dfference has occurred (Hll, Leghley, & Hnton-Andersson 995; Verba, Schlozman, & Brady, 995; Hll & Leghley, 996; Rngqust, Hll, Leghley & Hnton-Andersson, 997; Wnders, 999, Leghley, 2). In hs study of votng turnout n Boston durng presdental electons from 964 through 984, Burnham (987) also argued class dfferences were evdent n the declne of voter turnout. Hs research showed that voter turnout among blue collar workers dropped from 66. percent n 964 to 48 percent n 98 whle the turnout rate of whte collar workers dropped from 83.2 percent to 73 percent. Reter (979) also argued that the class bas n votng turnout has ncreased. Based on NES data from 96 to 976, Reter used ncome as a measure of class and found that the dfference between the votng turnout rates of the top and bottom ncome quartles has ncreased from 8.2 percent n 96 to 28.7 percent n 976. Usng educaton as a measure of class, Bennett (99) also argues that lower class ndvduals not only show lower rates of votng but that they also contrbuted to the declne of overall votng turnout. Hs fndngs showed that the voter turnout of whtes who had a college educaton dropped from 7 percent n 964 to 59 percent n 988. On the other hand, the turnout rate among ndvduals wthout a college educaton dropped from 63 percent to 3 percent. Although many of the works revewed above clearly fnd a relatonshp between soco-economc class and votng turnout, none of them managed to answer why. Pven and Cloward (2), however, shed some lght on ths queston. Not only dd they fnd that the lower class had the greatest turnout declne from 964 to 98, they also found out why the lower class declne was greater than for other groups. Emphaszng that the lower class had the potental to become strong Democratc partsans, Pven and Cloward found that the mechansm that kept voter turnout rates low for these ndvduals was lnked to the Democratc Party s moblzaton practces. Accordng to these authors, the Democratc Party s falure to moblze new voters was responsble for turnout declne among the lower class. They argued that regstraton fell among the lower class more than others due to less moblzaton by Democratc Party; consequently, ths led to the lower turnout among the lower class (Pven & Cloward, 2). Leghley and Nagler (992), however, ndcated that Bennett's measurement of educaton dd not accurately capture class snce the meanng of a college educaton n 964 was qute dfferent from n 988. Furthermore, t was more common n 964 for ndvduals not to have a college educaton. On the other hand, some scholars have challenged the concluson that class dfferences played a role n the declne of voter turnout n the Unted States. Usng varous demographc data from the NES and CPS, Leghley and Nagler (992) found that the class dfferences between voters and nonvoters n presdental electons reman the same from 964 through 988. They argued that although there was a wde range n turnout across ncome groups, ths fact dd not mply that there was a sgnfcant change n relatve turnout rates over tme. Instead, Leghley and Nagler found that, when focusng on ncome as a measure of class, the turnout rate of the lowest ncome groups dropped by 6 percent whle the rate of the hghest ncome groups dropped by 9. percent. Ths lack of dfference between the classes was also found by Shelds and Godel (997). In ther study on congressonal electons and class bases n voter turnout, they found that declnng voter turnout rates have occurred among all segments of socety snce the early 96s, not just among the lower classes. The New Amercan Voter (Mller & Shanks 996), drectly nspred by The Amercan Voter (Campbell, Converse, Mller & Stokes, 96), gves researchers thoughtful nformaton about voter turnout. Mller and Shanks (996) noted that t s puzzlng that voter turnout rates have declned whle educatonal levels have ncreased and voter regstraton laws have become moderated. However, they contend that generatonal dfferences accounted for these lower turnout rates. Dvdng voters nto two separate groups, pre-new Deal and post-new Deal, they argued that the turnout varaton had been explaned not by soco-economc factors, but rather by psychologcal poltcal nvolvement n dfferent areas. In ther research, the pre-new Deal group showed a very hgh nterest n poltcs whle the post-new Deal generaton was not as nvolved n poltcs. It s wdely accepted that economc condtons are assocated wth turnout. Wolfnger and Rosenstone (98) argued that rch ndvdual have a bgger stake n the system and are more hghly motvated to make the approprate choce on Electon Day and to support the poltcal system by partcpatng n t. Hll and Leghley (996) noted that lower class turnout can be ncreased when the economc condton of the state s stable, whle the economc recesson demoblzes the lower class. Radclff (992) also argued that when economc condtons are poor, voters would be demoblzed. Accordng to Radclff, when economc condtons are poor, people tend to skew ther attenton to personal Indana Journal of Poltcal Scence, Wnter 27, 7

3 concerns. Consequently they wthdraw from the poltcal process. Radclff also argued that under poor economc condtons, uneducated and lower ncome people tend to be affected by macroeconomc condtons at hgher degree than other people. On the other hand, people are not affected by poor economc condtons f economc securty programs are wellfunded (Radclff, 992). CONCEPTUALIZATION OF CLASS As a matter of fact, the conceptualzaton of class has not been easy. As revewed before, there are many ways to defne dfferent classes, varyng from educatonal level, occupatonal locaton, ncome, and collar types. Manza and Brooks (999) dd an excellent job of summarzng the varous types of class defntons. Accordng to them, the votng behavor lterature has typcally defned class n one of three dstnct ways. The most common approach s to dstngush between blue-collar and whte-collar workers. The assumpton behnd ths conceptualzaton of class s between the mddle class as a whole and the lower or workng class (Manza & Brooks, 999, p. 55). Despte ts popularty, defnng class based on collar types has some problems. Manza and Brooks (999, p. 55) wrte that: It s relatvely easy to see the lmtatons of such an assumpton. Frst, there are mportant sources of class dvsons wthn both the mddle class(es) and the workng class(es) that cannot be dentfed wth a two-class mode. For example, t s very dffcult to place routne whte-collar employees workng n servce ndustres. Whle such workers do not have manual employment, they hardly enjoy the benefts of the employment relatons typcal of professonal or manageral occupatons. Further, mportant changes n the class structures of captalst socetes snce World War II are dffcult to dentfy wth such a model. The second approach, accordng to Manza and Brooks (999), s most often used by contemporary socologsts, but s not commonly used n votng studes. Ths approach defnes class n terms of occupatonal locaton and/or employment stuaton. They contend that ths approach contans two dfferent conceptons of class, gradatonal and relatonal. The latter approach s more wdely accepted n the study of votng behavor. As stated by Manza and Brooks (999, p. 56): In relatonal approaches, dfferent clusters of occupatons are vewed as havng smlar though not dentcal employment stuatons and/or lfe chances. Rather than generatng a scale of all occupatons, relatonal approaches defne classes n terms of ether market or producton relatons. The TABLE. Geographcal Unts of Sample States State Aggregaton Type Number of Observatons Calforna Precnct 73 County 2 Mchgan Mnor Cvl Dvson 493 County Mnnesota Mnor Cvl Dvson 267 Mnor Cvl Dvson 43 wth Mult County County 2 New York County Subdvson 989 County 5 result s a set of categorcal dstnctons among actors based on ther employment stuaton. The thrd approach s to dstngush classes on the bass of an ndvdual s ncome. The basc logc behnd ths approach s straghtforward. Manza and Brook (999, p. 56) argue that [h]gher-ncome people have dfferent materal nterests than lower-ncome people. They are better able to fend for themselves n the market, and thus should have much less use for government-provded socal provson or progressve taxaton. Conversely, lower-ncome people should be expected to have the opposte nterests. However, Manza and Brooks (999) argue that there s a problem wth ths approach, statng that ndvduals wth the same level of ncome mght have dfferent long-term economc nterests. For example, a sem-sklled factory worker and a college student who s employed part-tme as a computer programmer mght report the same ncome, but ther expectatons n the long term mght be qute dfferent. Despte ths problem, there s enough justfcaton to conceptualze the class wth ncome. Leghley and Nagler (992) argued that ncome s more preferable to occupaton as the relevant measure of socoeconomc status because of three reasons. Frst, ncome s the more relevant measure wth regard to government polcy. Second, some occupatons are dffcult to categorze as whte- or blue-collar jobs. Fnally, occupatonal rankngs may not be stable over tme. DATASET AND METHODOLOGY In ths research, I estmate ndvdual lower class votng turnout n the 2 presdental electon for four states: Calforna, Mchgan, Mnnesota, and New York. Indana Journal of Poltcal Scence, Wnter 27, 7

4 TABLE 2. Dfference between Federal Electon Commsson and Merged Data Votng Age Populaton Votng Turnout Rate States Year FEC Report Merged Data FEC Report Merged Data Dfference Calforna 2 24,873, 24,558,5 44.% 45.3%.2% Mchgan 2 7,358, 7,329,6 57.5% 57.66%.6% Mnnesota 2 3,547, 3,857, % 63.75% - 5.5% New York 2 3,85, 4,279, % 48.82% -.58% Electon returns are electroncally archved n one of two locatons dependng upon the state: () on the ndvdual state s Secretary of State s webste or (2) on a major unversty webste for that state. Ths data s at the MCD, precnct level, or subcounty level.. Calforna was useful snce ts geographcal aggregaton s qute low at the precnct level. Mnnesota was added to the sample for comparson purposes as ts regstraton requrements are farly moderate when compared to other states. The selecton of Mchgan and New York was based solely on data avalablty at the tme of the research. I make no clam that ths small number of states can be generalzed for the entre Amercan electorate. Mergng Data: The electoral varables (.e., electon returns) were merged wth avalable demographc data publshed by the U.S. Census Bureau. Mergng the electoral data wth census nformaton requred several steps. Frst, the electoral nformaton (such as electon returns and total regstraton) was saved wth the geographcal dentfcaton. Second, the census nformaton wth the geographcal dentfcaton was saved at the correspondng level of aggregaton wth the electoral nformaton. Thrd, both census nformaton and electoral nformaton were merged together based on ther geographcal dentfcaton. At ths stage, some geographcal unts had to be aggregated due to the characterstcs of the geographc unts. Measurement Errors: When electoral data are merged wth the census nformaton, measurement errors are nevtable. After mergng the census data wth the 2 electon returns, some precncts, MCDs, and county subdvsons had to be excluded because the number of total votes exceeded the total votng age populaton. Ths happened because the census was not conducted at the same tme that the electon occurred. Despte ts ntrnsc errors, there are relatvely small errors exst for 2 data due to the short gap between electon date and census date. Table 2 shows the exstng measurement errors n each state. Varables and Data Collecton: The varables used n ths research ncluded: votng age populaton, unemployment rate, number of below poverty thresholds, total votes for presdental canddate n the general electons, and total regstraton. Votng Age Populaton. The votng age populaton s defned as the number of people who are 8 years or older. The votng age populaton for 2 s obtaned from 2 Census. 2 Lower Class. The operatonal defnton of class s based on the amount of ncome a partcular famly makes a year. Snce poverty status s used by federal agences n ther statstcal work to mplement ad programs, currently the most sutable way to defne a class s to look at the poverty status. In ther study enttled Lower-Class Moblzaton and Polcy Lnkage n the U.S. States, Hll, Leghley, and Hnton- Andersson (995) collected a number of ndcators to produce an ndex of class status, ncludng Jackman and Jackman s (983) occupatons, ncome levels, and Duncan s Socoeconomc Index scores. Combnng all of the data together, Hll, Leghley and Hnton- Andersson found that ther ndex was qute close to federal estmates of ndvduals lvng n poverty. The authors clamed that ther measure of lower class turnout s vald (Hll, Leghley, & Hnton-Andersson, 995, p. 78). Table 3 descrbes the poverty thresholds for 2 provded by the U.S. Census Bureau. 3 The number of persons who are 8 years or more and below the poverty level s calculated based on these census varables for each year. Usng ths method, I have calculated the sze of the lower class n for the four states under study. Ths data can be seen n Table 4. Throughout the states, the sze of the lower class s farly small consstng of around percent of the entre populaton. Mnnesota has the smallest lower class at 7.3 percent of the state s total populaton compared to 2 percent n Calforna and 2.4 percent n New York. Votng Turnout. The number of total votes for presdental canddates n the 2 electon s obtaned from each state s electon result table. Indana Journal of Poltcal Scence, Wnter 27, 72

5 TABLE 3. Poverty Thresholds n 2, by sze of Famly & Number of Related Chldren Under 8 (Dollars) Related chldren under 8 years Sze of Famly Unt Weghted Average Thresholds None One Two Three Four Fve Sx Seven 8 or more One person 8,794 Under 65 years 8,959 8, years and over 8,259 8,259 Two persons,239 Householder under 65 years,59,53,869 Householder 65 years and over,49,49,824 Three persons 3,738 3,47 3,86 3,874 Four persons 7,63 7,76 8,52 7,463 7,524 Fve persons 2,89 2,49 2,73 2,65 2,55 2,236 Sx persons 23,528 24,636 24,734 24,224 23,736 23,9 22,579 Seven persons 26,754 28,347 28,524 27,94 27,489 26,696 25,772 24,758 Eght persons 29,7 3,74 3,984 3,48 3,94 3,88 29,279 28,334 28,93 Nne persons or more 35,6 38,38 38,322 37,83 37,385 36,682 35,76 34,84 34,625 33,29 Source: U.S. Bureau of the Census, Current Populaton Survey. Unemployment Rate. Accordng to the Bureau of Labor Statstcs and the U.S. Census Bureau, unemployed persons can be defned as all persons who had no employment durng the reference week, were avalable for work, except for temporary llness, and had made specfc efforts to fnd employment some tme durng the 4-week-perod endng wth the reference week (Jont Project between the Bureau of Labor Statstcs and the Bureau of the Census, 996). The Census Bureau provdes the employment status for the populaton 6 years and over by gender. 4 The unemployment rates for each state are provded n Table 5. In 22, New York had the lowest unemployment rate (3.6 percent) of the four states, compared to 7.5 percent n Calforna and 6.2 percent n Mchgan. Snce the U.S. Census only provdes average unemployment rates for the prevous year, t s not possble to observe the amount of change n the unemployment rate whch may better capture the economc fluctuatons relevant to voters. TABLE 4. Sze of the Lower Class by State, 2 State Lower Class Non-Lower Class Calforna.2.88 Mchgan.9.99 Mnnesota New York TABLE 5. Unemployment Rate by State* State Unemployment Rate Calforna 7.5% Mchgan 6.2% Mnnesota 4.9% New York 3.6% * 2 unemployment rate was calculated from the U.S. Census Summary Fle 3. Indana Journal of Poltcal Scence, Wnter 27, 73

6 A Descrpton of Ecologcal Inference: Gven data on the lower class and Presdental votes n each observaton (MCD, sub county, or precnct), I turned to the task of estmatng turnout by class by usng ecologcal nference. The basc structure of ecologcal nference s that each observaton (MCD, subcounty, or precnct) s treated as a separate 2x2 table wth known margnals (number of lower class/non-lower class votng age populaton by number of voters/nonvoters) and wth unknown nner cells (the number of lower class voters). Table 6 shows the basc 2x2 structure of ecologcal nference used n ths research. N, N, Y, and N - Y are non-negatve ntegers that are observed representng the number of the lower class votng age populaton, number of non-lower class votng age populaton, number of voters, and number of non-voters n MCD. The nner cell entres Y and Y are not observed. Y represents the Number of the votes by the lower class and Y t represents the number of the votes by the non lower class. It s assumed that: Y o N Y ~ Bnomal( N, p N ~ Bnomal( N, p ) ) and where p, and p are ultmate quanttes of nterests representng the probablty of a vote by the lower class and the probablty of a vote by the non lower class respectvely. Snce the probablty that an ndvdual votes, q, s the weghted sum of two ndependent probabltes: the probablty of a vote by the lower class, p, and the probablty of a vote by non lower class, p, we can express the margnal probablty that an ndvdual votes as follow: q = p x + p ( x ) where, x = N / N and - x = N / N, are respectvely the observed proportons of the lower class and nonlower class. Ths basc structure of ecologcal nference and ts notatons wll be revsted often as I dscuss the Bayesan approach below. Bayesan Herarchcal Model. The basc model elaborated n Table 6, descrbes the observed total number of vote n the MCD as a draw from a bnomal dstrbuton wth parameters x and N. As ndcated n Table 6, the probablty that an ndvdual votes, x, s the weghed sum of two ndependent probabltes: the probablty of a vote by the lower class, p, and the probablty of a vote by non lower class, p. As ndcated prevously, ths step ntroduces an accountng dentty, the logcal boundares that are mpled by the data. Three peces of nformaton from the MCD data enters the lkelhood: the number of vote, V; the number of the lower class, L; and the number of non lower class, U. The probabltes of a vote by the lower class and non-lower class, p, and p are normalzed through TABLE 6. 2x2 Ecologcal Inference Table Y = Y = (non-vote) (vote) X = (lower class) Y N X = (non-lower class) Y N N - Y Y N transformaton to the logstc. The formal representtaton of ths process s straghtforward: p p V ~ bn( x, N ) q = p x + p ( x ) N = U + L x = L / (U + L ) = exp( θ ) /( + exp( θ )) = exp( θ ) /( + exp( θ )) The prmary assumptons of the modelng process are that the logt of the lower class turnout n each MCD observaton s drawn from a sngle underlyng normal dstrbuton and that the logt of non-lower class turnout n each MCD s drawn from a separate underlyng normal dstrbuton. Although there are a number of alternatve dstrbutonal assumptons to ths bnomal-normal model, scholars used ths strategy most often to estmate voter turnout n ecologcal nference applcatons (Corder & Wolbrecht, 24a). The herarchcal model can be used when nformaton s avalable on several dfferent levels of observatonal unts. In ths research, manly MCD level and state level observatons were avalable. The herarchcal structure of the model was ntroduced by specfyng the normal dstrbuton that descrbes the MCD logts. At the second stage, for each MCD level probablty, the pror dstrbuton s treated as a draw from a normal dstrbuton wth mean µ and varance 2 σ where: 2 2 θ ~ N( µ, σ ), θ ~ N( µ, σ ). θ s assumed to be a pror ndependent of θ for all s. In addton, we assume the followng hyperprors: µ N( m, ), ~ M ~ N( m, M µ ), Indana Journal of Poltcal Scence, Wnter 27, 74

7 TABLE 7. Proporton of Votng Turnout by Class State Observed Total Turnout Rate* Lower Class Posteror Medan Lower Class 95% BCI Non-Lower Class Posteror Medan Non-Lower Class 95% BCI Mchgan [ ].6 [ ] Mnnesota [ ].657 [ ] New York [ ].589 [ ] Calforna [ ].478 [ ] * Observed turnout rate could be dfferent from the actual turnout rate reported from Secretary of State Offce due to measurement errors as dscussed n the earler secton. 2 σ ~ IG ( / 2, σ / 2) σ ~ IG / 2, / 2) 2 ν, ( σ ν. Ths hyper pror state-level mean and varance are specfed n a way that adds no nformaton to the model. A unform dstrbuton or relatvely flat normal dstrbuton centered around zero would be approprate for the mean. A smlarly flat pror would be ntroduced for the varance. Accordng to Corder and Wolbrecht (24b), the Bayesan strategy both ncorporates nformaton about and permts a test of the mpact of contextual factors. The probablty of a vote by the lower class and non lower class and the regresson coeffcents for the contextual effects were smultaneously estmated for the entre set of avalable data at each electon year. In order to estmate the contextual effects, t s addtonally assumed that the probablty of a vote by the lower class s condtonal on a contextual varable, whch allows the probablty of a vote by lower class and non lower class vary over MCDs accordng to the unemployment rate, covarate employed n ths research (Kng 997). After a vector of acceptable canddate values for the probablty of a vote by the lower class and the probablty of a vote by the non lower class are selected, the bnomal probablty of a vote by lower class and non-lower class are transformed va the logstc. The logts are ndependently regressed on the contextual factor. Once we obtan the vector of populaton-weghted lnear regresson coeffcents, the vectors were retaned and used n the calculaton of the lkelhood n the subsequent teraton of the model (Corder & Wolbrecht 24b). Once they enter the lkelhood, new canddate values were selected, regresson coeffcents were updated, and ths process was repeated. Canddate values and regresson parameters were updated va Markov Chan Monte Carlo (Corder & Wolbrecht, 24b). Markov Chan Monte Carlo methods are mplemented n the MCMCpack, an R package authored by Martn and Qunn (23). I also used a modfed MCMCpack R package authored by Corder and Wolbrecht (24b) to mplement constraned models (the probablty of a vote by the lower class would not exceed the probablty of a vote by nonlower class). Startng from an unnformed pror, an MCMC smulaton terates a number of tmes to converge on model solutons. For ths research each smulaton was 7, teratons wth the frst 5, teratons dscarded as the burn-n. The medan from each montored chan was treated as a pont estmate for each quanttes of nterest. The pont estmates are reported n the next secton of the artcle. In addton, the tables nclude the regon of 95 percent hghest posteror densty, also known as the Bayesan Credble Interval (BCI). RESULTS AND DISCUSSION Turnout Estmates: An examnaton of the turnout model resulted n a number of fndngs whch have been summarzed n Table 7. The estmates suggest that there was a sgnfcant dfference between lower class partcpaton and non-lower class partcpaton. In most states, the voter turnout of the lower class s lower than that of the non-lower class. The estmaton of the probablty of a vote by lower class was consstently lower than the probablty of a vote by non-lower class. Dagnoss of the Turnout Model: The emprcal results from a gven MCMC analyss may not be relable untl the chan has reached ts statonary dstrbuton. A possble way to see the model s performance s to observe convergence propertes. In ths secton, the Cty of Kalamazoo was used to observe convergence Indana Journal of Poltcal Scence, Wnter 27, 75

8 TABLE 8. Hedelberger and Welch Test for Selected MCDs n States Statonarty test Halfwdth Test Mean Halfwdth Mchgan Lower Class Passed Passed Non-Lower Class Passed Passed Mnnesota Lower Class Passed Passed Non-Lower Class Passed Passed New York Lower Class Passed Passed Non-Lower Class Passed Passed Calforna Lower Class Passed Passed Non-Lower Class Passed Passed.36.7 propertes for the pont estmate of the probablty of a vote n 2 presdental electon. If the posteror dstrbuton s not converged, multmodalty of the posteror densty would be shown; ths s a classc sgn of nonvergence (Gll 22). Fgure [See appendx] revealed no multmodalty for ether lower class or non-lower class posteror dstrbuton. It was also evdent that the posteror dstrbuton for nonlower class turnout was narrowly focused around the pont estmate whle the posteror dstrbuton for lower class turnout was a bt wder. Ths narrow posteror dstrbuton s another ndcator of the model s convergence. Fgure 2 and 3, whch are ncluded n the appendx to ths artcle, ndcate that the model has converged. Fgure 2 shows the horzontally flat trend lne ndcatng that there s no vsble trend n the smulaton. Fgure 3 shows no sgn of autocorrelaton n the posteror dstrbuton. Both of these fgures whch represent the propertes of the convergence show that the smulated chan s statonary. The Hedelberger and Welch dagnostc was used to see f the estmates were drawn from a chan that has converged. MCDs from Mchgan and Mnnesota, a subcounty dvson from New York, and a precnct from Calforna were arbtrarly selected to test convergence. It s vrtually mpossble to test convergence for all unts. Table 8 reports the results of the dagnostc. In ths table, t s farly obvous that the models pass the basc convergence dagnoss. Based on the graphcal dagnoss of the convergence and the Hedelberger and Welch test, t can be concluded that the pont estmates were drawn from a chan that has converged. Table 9. Correlaton Coeffcents between Lower Class and Non-Lower Class* State Correlaton Coeffcents Calforna.63 Mchgan.68 Mnnesota.7 New York.56 * Correlaton coeffcents are calculated by STATA 7. for wndows based on pont estmates of the lower class turnout and non lower class turnout Effects of the Unemployment Rate on the Probablty of a Vote: One smple way to see f there s any dfference between the lower class and the non-lower class n respondng to external stmul s to observe how the lower class turnout vares wth non-lower class turnout. Table 9 shows the correlaton coeffcents between the lower class turnout and non-lower class turnout n the varous states. The hghest correlaton coeffcent was observed among the Mnnesota voters (r =.7) and the lowest was found among the New York voters (r =.56). The correlaton coeffcents between the lower class turnout and non-lower class turnout n Mchgan were moderate; ths suggested that both classes response to contextual stmul to some degree. The lower class n New York responded to contextual stmul dfferently from the non-lower class whch s shown by the correlaton coeffcent (r=.56). Ths smple observaton confrms the ntal assumpton that each class responds to contextual stmul dfferently. For the 2 presdental electon, the regresson coeffcents were obtaned by usng the Bayesan Indana Journal of Poltcal Scence, Wnter 27, 76

9 Table. Explanng Varaton n Lower Class and Non-Lower Class Turnout by Unemployment Rate Calforna Mchgan Mnnesota New York Lower Class Regresson Coeffcent [ Constant [ ] Non Lower Class Regresson Coeffcent [ ] Constant.422 [ ] -2.8 [ ] -.34 [ ] [ ].753 [ ] [ ] -.72 [ ] -.39 [ ].75 [.4 -.8] -7. [ ] [ ] [ ].697 [ ] Notes: Dependent varable: logt of the proporton of age elgble populaton castng votes for Presdent. Estmated va Markov Chan Monte Carlo. 95% Bayesan Credble Interval (BCI) n brackets. Herarchcal model. Each table shows the estmates from each state by class. The Bayesan Credble Intervals (BCI) ndcated that the values of the hghest posteror densty regons coverng the 95 percent of the posteror dstrbuton wth the hghest probablty (Gll, 22). If zero les n the BCI, then the effect of the contextual effect s trval. If zero s not n the BCI, then the unemployment rate provdes nformaton about MCD level varaton n turnout. Effects of Unemployment. Table shows the regresson coeffcents of the unemployment rate n each state. In 2, the lower class n Mnnesota and New York dd respond to economc condtons. BCIs exclude zero. Once the unemployment rate ncreased, the probablty of the lower class turnout decreased. The Mchgan, Mnnesota, and New York cases all showed that the unemployment rate s negatvely related to the probablty of non-lower class turnout. The drecton of the functon was consstently downward, whch ndcated that the hgher the unemployment rate, the lower the probablty of that the non-lower class wll vote. We can see the effects of the unemployment rate on the probablty of lower class and non-lower class turnout n a graphcal way by ntroducng regresson coeffcents and constants nto the specfcaton of the logt model: E( Y ) = + e α β where E(Y) s the probablty of a vote by the lower class, α and β are constant and regresson coeffcent respectvely, and X s the unemployment rate for each MCD. Fgures from Fgure 4 through Fgure 7, whch are ncluded n the appendx, show the same characterstcs: the non-lower class responded more sharply to the X unemployment rate change than the lower class dd. Ths also s an expected result. Overall, the estmates ndcate non-lower class voters respond strongly to macroeconomc condtons. Lower class voters do not. Dagnoss. Table reports the results of the Hedelberger and Welch Test. As Table demonstrates, all regresson coeffcents are drawn from statonary whch suggests that the nterpretatons and analyses based on these pont estmates are relable. CONCLUSION When the Natonal Voter Regstraton Act, more commonly known as the Motor Voter Act, was passed n 993, hgher regstraton and consequently hgher votng turnout were naturally expected. The frst goal seemed to be acheved when the Federal Electon Commsson reported at least mllon newly regstered ndvduals after law s passage (FEC, 997). Votng turnout, however, dd not follow the same drecton. The turnout rate n the frst presdental electon after the Motor Voter Act was only 49 percent the lowest turnout rate snce World War II. The Natonal Voter Regstraton Act was just the most recent effort to make regstraton easer. Snce the 96s, regstraton has become sgnfcantly easer. There s, however, a fundamental puzzle when observng overall voter turnout levels n Amercan presdental electons. Despte varous efforts to reduce the barrers to votng, ndvduals do not vote as much as they dd before. Who votes and who does not? These are the endurng questons n Amercan poltcs. Ths research quanttatvely tested two hypotheses that provde answers to these central questons along wth effects of contextual factors on the turnout. Frst, I hypotheszed that lower class voter turnout s lower than non-lower class voter turnout. Second, turnout Indana Journal of Poltcal Scence, Wnter 27, 77

10 TABLE. Hedelberger and Welch Test for Coeffcents n States Statonarty Test Halfwdth Test Mean Halfwdth Mchgan Lower Class Passed Passed Non-Lower Class Passed Passed Mnnesota Lower Class Passed Passed Non-Lower Class Passed Passed New York Lower Class Passed Passed Non-Lower Class Passed Passed Calforna Lower Class Passed Passed Non-Lower Class Passed Passed among the lower classes s less responsve to economc condtons than the non-lower class. Two Bayesan herarchcal ecologcal nference methods were employed to test these hypotheses: the basc and extended models. Fndngs: The results of ths research confrmed the frst hypothess. The probablty of a vote by the lower class s lower than that of the non-lower class. As dscussed prevously n ths artcle, the lower class n all states voted at lower rates than dd the non-lower class. Furthermore, the probabltes of a vote by the lower class are statstcally dfferent from those by the nonlower class. Ths result s consstent wth the fndngs of prevous researchers. The most valuable advantage of havng a covarate n the model s that we can estmate the effects of a contextual factor on votng turnout. In ths artcle, the second hypothess dealt wth the macroeconomc context. In general, I found that the responsveness of lower class voter turnout was dfferent from that of the non-lower class. That s, voter turnout for the nonlower class decreased as unemployment ncreased. Ths effect was much weaker for the lower class. These results confrmed Radclff s (992) concluson that the voters would be demoblzed when macroeconomc condtons were poor; however, ths research does not support Radclff s concluson that voter turnout among the lower classes would be most affected. Ths may be explaned by the exstence of economc securty programs. Radclff (992) argued that when economc securty s provded, the degree of demoblzaton would be smaller although macroeconomc condtons are poor. Snce the lower class, n ths research, was defned based on the poverty status and these ndvduals tend to receve welfare benefts, they would not be demoblzed as much as non-lower class. Contrbutons: Ths research confrmed and strengthened the prevous conclusons about Amercan votng behavor by usng a relatvely new method, Bayesan herarchcal ecologcal nference. The artcle demonstrated how economc class s assocated wth voter turnout. Furthermore, t ncreased our knowledge of how macroeconomc condtons affect votng turnout among people from dfferent economc standngs. If, as I confrmed, the low turnout rate among the lower class s persstent, these ndvduals would have less of a chance to mpact publc polcy. In the end, the prmary accomplshment of ths research has been to reveal how and to what extent economc class and macroeconomc condtons are assocated wth votng turnout n the Unted States. NOTES Recently some scholars rased a queston of usng the votng age populaton to calculate the turnout rate (McDonald & Popkn, 2; Martnez, 23). Accordng to them, the Votng Elgble Populaton should be used to calculate turnout rate nstead of Votng Age Populaton because the VAP ncludes nonctzens and felons who are not elgble to vote. Due to data avalablty, the VAP s used n ths research. Gven the fact that the VAP s used, ecologcal nference models mght underestmate lower class turnout snce the VEP s almost dentcal wth Indana Journal of Poltcal Scence, Wnter 27, 78

11 VAP for nonlower class whle the VEP tends to be smaller than the VAP for lower class. 2 Census 2 Summary Fle 3 provdes sex by age at P8. For votng age populaton, both the male and female 8 years and over populaton are extracted and calculated. 3 The Bureau of Census provdes the prevous year s poverty status by age n P87 for 2 census. 4 P43 for 2 Census Summary Fle 3 provdes the number of people n labor force, employed, and unemployed for both gender. REFERENCES Bennett, S. E. (99). Left behnd: Explorng declnng turnout among noncollege young whtes, Socal Scence Quarterly, 72, Boyd, R. W. (98). Declne of U.S. voter turnout: Structural explanatons. Amercan Poltcs Quarterly, 9, Burnham, W. D. (987). The turnout problem. In Schlozman, K. L. (ed.). Electons Amercan Style. Washngton, D.C.: Brookngs. Campbell, A., Converse, P. E., Mller, W. E., & Stokes, D. E. (96). The Amercan voter. New York: Wley. Cassel, C. A. & Hll, D. B. (98). Explanatons of turnout declne: A multvarate test. Amercan Poltcs Quarterly, 9, Corder, J. K. & Wolbrecht, C. (24a). Usng pror nformaton to ad ecologcal nference: A Bayesan approach. In Kng, G., Rosen, O. & Tanner, M. (eds.). Ecologcal nference: New Methodologcal Strateges. New York: Cambrdge Unversty Press. Corder, J.K. & Wolbrecht, C. (24b). Incorporatng women voters after suffrage. Forthcomng publcaton. Federal Electon Commsson (n.d.). Statstcal Hghlghts of the Federal Electon Commsson Report to the Congress on the Impact of the Natonal Voter Regstraton Act Retreved January, 25, from Gll, J. (22). Bayesan methods: A socal and behavoral scences approach. New York: Chapman & Hall/CRC. Hll, K.Q. & Leghley, J.E. (996). Poltcal partes and class moblzaton n contemporary Unted States electons. Amercan Journal of Poltcal Scence, 4, Hll, K.Q., Leghley, J.E. & Hnton-Andersson, A. (995). Lower-class moblzaton and polcy lnkage n the U.S. states. Amercan Journal of Poltcal Scence, 39, Jackman, M. R. & Jackman, R. W. (983). Class awareness n the Unted States. Berkeley, CA: Unversty of Calforna Press. Kng, G. (997). A soluton to the ecologcal nference problem: Reconstructng ndvdual behavor from aggregate data. Prnceton, N.J.: Prnceton Unversty Press. Kornbluh, M. L. (2). Why Amerca stopped votng: The declne of partcpatory democracy and the emergence of modern Amercan poltcs. New York: New York Unversty Press. Leghley, J. E. & Nagler, J. (992). Socoeconomc class bas n turnout, : The voters reman the same. Amercan Poltcal Scence Revew, 86(3), Leghley, J. E. (2). Strength n numbers? The poltcal moblzaton of racal and ethnc mnortes. Prnceton, M.J.: Prnceton Unversty Press. Manza, J. & Brooks, C. (999). Socal cleavages and poltcal change: Voter algnments and U.S. party coaltons. New York: Oxford Unversty Press. Martn, A. D. & Qunn, K. M. (23). MCMCPack. A package dstrbuted by the R project for Statstcal Computng. Last retreved December 5, 27, from Martnez, M. D. (23). Comments on Voter turnout and the Natonal Electon Studes. Poltcal Analyss,, McDonald, M. P. & Popkn, S. L. (2). The myth of the vanshng voter. Amercan Poltcal Scence Revew, 95, Mlbrath, L. W. & Goel, M. L. (977). Poltcal partcpaton. Chcago: Rand McNally. Mller, W. E. & Shanks, J. M. (996). The New Amercan voter. Cambrdge, MA: Harvard Unversty Press. Patterson, T. E. (22). The vanshng voter: Publc nvolvement n an age of uncertanty. New York: Random House. Pven, F. F. & Cloward, R. A. (2). Why Amercans stll don t vote: And why poltcans what t that way. Boston, MA: Beacon Press. Radclff, B. (992). The welfare state, turnout, and the economy: A comparatve analyss. Amercan Poltcal Scence Revew, 86, Indana Journal of Poltcal Scence, Wnter 27, 79

12 Reter, H. L. (979). Why s turnout down. Publc Opnon Quarterly, 43, Rngqust, E.J., Hll, K.Q., Leghley, J.E. & Hnton- Andersson, A. (997). Lower-class moblzaton and polcy lnkage n the U.S. states: A correcton. Amercan Journal of Poltcal Scence, 4, Shaffer, S. D. (98). A multvarate explanaton of decreasng turnout n presdental electons Amercan Journal of Poltcal Scence, 25, Shelds, T. G. & Godel, R. K. (997). Partcpaton rates, socoeconomc class bases, and congressonal electons: A crossvaldaton. Amercan Journal of Poltcal Scence, 4, Texera, R. A. (987). Why Amercans don t vote: Turnout declne n the Unted States New York: Greenwood. Verba, S., Schlozman, K. L. & Brady, H. E. (995). Voce and equalty: Cvc voluntarsm n Amercan poltcs. Cambrdge, MA: Harvard Unversty Press. Wnders, B. (999). The roller coaster of class conflct: Class segments, mass moblzaton, and voter turnout n the U.S., Socal Forces, 77(3), Wolfnger, R. E. & Rosenstone, S. J. (98). Who votes? New Haven: Yale Unversty Press. Indana Journal of Poltcal Scence, Wnter 27, 8

13 APPENDIX FIGURE. Posteror Dstrbuton of the Fracton of Lower Class and Non-Lower Class Turnout: Kalamazoo Cty, Mchgan, 2 FIGURE 2. Trace of Smulated Values of Lower Class and Non-Lower Class Turnout: Kalamazoo Cty, Mchgan, 2 Indana Journal of Poltcal Scence, Wnter 27, 8

14 FIGURE 3. Posteror Autocorrelaton, Non-Lower Class and Lower Class Turnout: Kalamazoo Cty, Mchgan, 2 FIGURE 4. Probablty of a Vote by the Unemployment Rate: Mnnesota 2 Presdental Electon All Turnout Non Lower Class Turnout Lower Class Turnout.5 Unemployment Rate Indana Journal of Poltcal Scence, Wnter 27, 82

15 FIGURE 5. Probablty of a Vote by the Unemployment Rate: Mchgan 2 Presdental Electon All Turnout Non Lower Class Turnout Lower Class Turnout.5 Unemployment Rate FIGURE 6. Probablty of a Vote by the Unemployment Rate New York 2 Presdental Electon All Turnout Non Lower Class Turnout Lower Class Turnout.5 Unemployment Rate Indana Journal of Poltcal Scence, Wnter 27, 83

16 FIGURE 7. Probablty of a Vote by the Unemployment Rate Calforna 2 Presdental Electon All Turnout Non Lower Class Turnout Lower Class Turnout.5 unemp Indana Journal of Poltcal Scence, Wnter 27, 84